The notebook for NaN Analysis post (https://dev.to/sephib/nans-bites-17kk).

nans_bites.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Background\n",
    "While working on a health-related classification project we encountered a very large [sparse metrix](https://en.wikipedia.org/wiki/Sparse_matrix), due to the vast amount of health/lab tests that were available. After a meeting with the business domain experts, we understood that our initial data preprocessing for removing missing data (NaN) was faulty.  \n",
    "In this post, we would like to share the pit-fall that we experienced and share our process for identifying features with missing values related to classification problems.  \n",
    "\n",
    "# The Pit Fall  \n",
    "We had over 2K of features across 40K patients. We knew that most of the features had a significant amount of `NaN`s, so we used the common methods - such as  [scikit-learn's VarianceThreshold](https://scikit-learn.org/stable/modules/generated/sklearn.feature_selection.VarianceThreshold.html) and [caret's near zero variance](http://topepo.github.io/caret/pre-processing.html#zero--and-near-zero-variance-predictors) functions to remove features with high `NaN` values.  \n",
    "We were left with less than 70 features and ran our base model to see if our classifier model could predict better than randomness. After displaying the [feature importance](https://catboost.ai/docs/concepts/python-reference_catboostclassifier_get_feature_importance.html#python-reference_catboostclassifier_get_feature_importance) from our `CatBoost` model, some concerns were raised regarding some of the features.  \n",
    "So we went back and did some homework...  \n",
    "\n",
    "While re-analyzing the features that were left, we saw that although they had passed \n",
    "our initial `NaN` tests, we did not check for the distribution of the NaN across our classes, i.e. some features had a significant amount of NaNs concentrated in a specific class which were not evenly distributed across our group classifications."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Data sample  \n",
    "To demonstrate this process let's look at an example dataset  - the [Horse Colic](https://archive.ics.uci.edu/ml/datasets/Horse+Colic) dataset.  \n",
    "This dataset includes the outcome/survival of horses diagnosed with colic disease based upon their past medical histories.   "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "df.shape: (299, 28)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>surgery</th>\n",
       "      <th>Age</th>\n",
       "      <th>Hospital_Number</th>\n",
       "      <th>rectal_temperature</th>\n",
       "      <th>pulse</th>\n",
       "      <th>respiratory_rate</th>\n",
       "      <th>temperature_of_extremities</th>\n",
       "      <th>peripheral_pulse</th>\n",
       "      <th>mucous_membranes</th>\n",
       "      <th>capillary_refill_time</th>\n",
       "      <th>...</th>\n",
       "      <th>packed_cell_volume</th>\n",
       "      <th>total_protein</th>\n",
       "      <th>abdominocentesis_appearance</th>\n",
       "      <th>abdomcentesis_total_protein</th>\n",
       "      <th>outcome</th>\n",
       "      <th>surgical_lesion</th>\n",
       "      <th>type_of_lesion1</th>\n",
       "      <th>type_of_lesion2</th>\n",
       "      <th>type_of_lesion3</th>\n",
       "      <th>cp_data</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ID</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>530101</td>\n",
       "      <td>38.50</td>\n",
       "      <td>66</td>\n",
       "      <td>28</td>\n",
       "      <td>3</td>\n",
       "      <td>3</td>\n",
       "      <td>NaN</td>\n",
       "      <td>2</td>\n",
       "      <td>...</td>\n",
       "      <td>45.00</td>\n",
       "      <td>8.40</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>died</td>\n",
       "      <td>2</td>\n",
       "      <td>11300</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>534817</td>\n",
       "      <td>39.2</td>\n",
       "      <td>88</td>\n",
       "      <td>20</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "      <td>...</td>\n",
       "      <td>50</td>\n",
       "      <td>85</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>euthanized</td>\n",
       "      <td>2</td>\n",
       "      <td>2208</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2 rows × 28 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   surgery  Age  Hospital_Number rectal_temperature pulse respiratory_rate  \\\n",
       "ID                                                                           \n",
       "0        2    1           530101              38.50    66               28   \n",
       "1        1    1           534817               39.2    88               20   \n",
       "\n",
       "   temperature_of_extremities peripheral_pulse mucous_membranes  \\\n",
       "ID                                                                \n",
       "0                           3                3              NaN   \n",
       "1                         NaN              NaN                4   \n",
       "\n",
       "   capillary_refill_time  ... packed_cell_volume total_protein  \\\n",
       "ID                        ...                                    \n",
       "0                      2  ...              45.00          8.40   \n",
       "1                      1  ...                 50            85   \n",
       "\n",
       "   abdominocentesis_appearance abdomcentesis_total_protein     outcome  \\\n",
       "ID                                                                       \n",
       "0                          NaN                         NaN        died   \n",
       "1                            2                           2  euthanized   \n",
       "\n",
       "   surgical_lesion type_of_lesion1 type_of_lesion2 type_of_lesion3 cp_data  \n",
       "ID                                                                          \n",
       "0                2           11300               0               0       2  \n",
       "1                2            2208               0               0       2  \n",
       "\n",
       "[2 rows x 28 columns]"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "from itertools import combinations\n",
    "\n",
    "pd.options.display.float_format = \"{:,.2f}\".format\n",
    "\n",
    "names = \"surgery,Age,Hospital Number,rectal temperature,pulse,respiratory rate,temperature of extremities,peripheral pulse,mucous membranes,capillary refill time,pain,peristalsis,abdominal distension,nasogastric tube,nasogastric reflux,nasogastric reflux PH,rectal examination,abdomen,packed cell volume,total protein,abdominocentesis appearance,abdomcentesis total protein,outcome,surgical lesion,type of lesion1,type of lesion2,type of lesion3,cp_data\"\n",
    "names = names.replace(\" \", \"_\").split(\",\")\n",
    "file_path = (\n",
    "    \"https://raw.githubusercontent.com/jbrownlee/Datasets/master/horse-colic.csv\"\n",
    ")\n",
    "df = pd.read_csv(file_path, names=names)\n",
    "label_col = \"outcome\"\n",
    "\n",
    "\n",
    "def preprocess_df(df):\n",
    "    df.columns = [c.strip() for c in df.columns]\n",
    "    df = df.replace(\"?\", np.nan).replace(\"nan\", np.nan)\n",
    "    df = df[~(df[\"outcome\"].isna())].copy()  # clean up label column\n",
    "    df.index.name = \"ID\"\n",
    "    df[label_col] = (\n",
    "        df[label_col]\n",
    "        .astype(str)\n",
    "        .replace({\"1\": \"lived\", \"2\": \"died\", \"3\": \"euthanized\"})\n",
    "    )\n",
    "    return df\n",
    "\n",
    "\n",
    "df = preprocess_df(df)\n",
    "print(f\"df.shape: {df.shape}\")\n",
    "df.head(2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First we will check how many featurs have `NaN` values."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The number features with NaN values are 19\n"
     ]
    }
   ],
   "source": [
    "# There are 28 features in this dataset\n",
    "def number_of_features_with_NaN(df):\n",
    "    _s_na = df.isna().sum()\n",
    "    return len(_s_na[_s_na > 0])\n",
    "\n",
    "\n",
    "print(f\"The number features with NaN values are {number_of_features_with_NaN(df)}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "## Near Zero Variance\n",
    "\n",
    "Simulating our intitial workflow, we will remove features with NaN values with our implementation of caret's R library [near zero variance](\"http://topepo.github.io/caret/pre-processing.html#zero--and-near-zero-variance-predictors\") function (with their default values). \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "After processeing the dataset via `near_zero_variance` we are left with 18 features with NaN values.\n",
      "\n"
     ]
    }
   ],
   "source": [
    "def near_zero_variance(df, cols=None, frq_cut=95 / 5, unique_cut=10):\n",
    "    if not cols:\n",
    "        cols = df.columns\n",
    "    drop_cols = []\n",
    "    for col in cols:\n",
    "        val_count = list(\n",
    "            df[col].value_counts(dropna=False, normalize=True).to_dict().items()\n",
    "        )\n",
    "        if len(val_count) == 1:\n",
    "            drop_cols.append(col)\n",
    "            continue\n",
    "        lunique = len(val_count)\n",
    "        percent_unique = 100 * lunique / len(df[col])\n",
    "        freq_ratio = val_count[0][1] / val_count[1][1] + 1e-5\n",
    "\n",
    "        if (freq_ratio > frq_cut) & (percent_unique <= unique_cut):\n",
    "            drop_cols.append(col)\n",
    "    return df[[c for c in df.columns if c not in drop_cols]]\n",
    "\n",
    "\n",
    "df_nzr = near_zero_variance(df)\n",
    "print(\n",
    "    f\"After processeing the dataset via `near_zero_variance` we are left with {number_of_features_with_NaN(df_nzr)} features with NaN values.\\n\"\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Deeper NaN  Analysis\n",
    "Since we are only interested in understanding the missing values in the dataset, we can view how many `NaN` values are in the various features.\n",
    "Let's now plot the remaining features relative to the percent of `NaN`s in each feature"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "def plot_percent_nan_in_features(df):\n",
    "    nan = df.isna().sum() / len(df)\n",
    "    feature_nan_threshold = {\n",
    "        round(threshold, 2): len(df.columns) - len(nan[nan < threshold])\n",
    "        for threshold in np.arange(0, 1.01, 0.05)\n",
    "    }\n",
    "    _df = pd.DataFrame.from_dict(\n",
    "        feature_nan_threshold, orient=\"index\", columns=[\"num_of_features\"]\n",
    "    )\n",
    "    _df.plot(\n",
    "        kind=\"bar\",\n",
    "        ylabel=\"Number of Features\",\n",
    "        xlabel=\"Percentage of NaNs in feature\",\n",
    "        figsize=(10, 5),\n",
    "        grid=True,\n",
    "    )\n",
    "\n",
    "\n",
    "plot_percent_nan_in_features(df_nzr)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "From the above plot, we can see that the number of features with more than 40% of NaNs are 2 features and above 25% are 6 features.  \n",
    "\n",
    "We can see that some of the features have very high `NaN` values.  \n",
    "\n",
    "Let's now remove these problematic features.\n",
    "For this example we will set a threshold of 35% . \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "After drop_features_above_threshold_max_na the number of features with NaNs that we are left with are : 14\n"
     ]
    }
   ],
   "source": [
    "threshold_max_na = 0.35\n",
    "\n",
    "\n",
    "def drop_features_above_threshold_max_na(df, threshold_max_na=threshold_max_na):\n",
    "    nan = df.isna().sum() / len(df)\n",
    "    nan_threshold = nan[nan > threshold_max_na]\n",
    "    df = df.drop(columns=nan_threshold.index).copy()\n",
    "    return df\n",
    "\n",
    "\n",
    "df_nzr_threshold = drop_features_above_threshold_max_na(df_nzr)\n",
    "print(\n",
    "    f\"After drop_features_above_threshold_max_na the number of features with NaNs that we are left with are : {number_of_features_with_NaN(df_nzr_threshold)}\"\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We may assume that we have removed the problematic features and can try to imputate our NaN data and run our pipeline/model.  \n",
    "\n",
    "But before we do so let's look a bit more closely at our classification. \n",
    "\n",
    "## NaN Distribution Among the Classifer Labels \n",
    "Looking at the classifier label feature `outcome` we can see the following distribution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>num</th>\n",
       "      <th>percent</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>lived</th>\n",
       "      <td>178</td>\n",
       "      <td>0.60</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>died</th>\n",
       "      <td>77</td>\n",
       "      <td>0.26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>euthanized</th>\n",
       "      <td>44</td>\n",
       "      <td>0.15</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            num  percent\n",
       "lived       178     0.60\n",
       "died         77     0.26\n",
       "euthanized   44     0.15"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def create_value_counts_df(df, s2=pd.Series(dtype=float), col=None, check_na=False):\n",
    "    if col:\n",
    "        s1 = df[col].value_counts(dropna=False)\n",
    "    if check_na:\n",
    "        s1 = df.isna().sum()\n",
    "        s2 = df.isna().sum() / len(df)\n",
    "    if s2.empty:\n",
    "        s2 = df[col].value_counts(normalize=True, dropna=False)\n",
    "    df = pd.concat([s1, s2], axis=1)\n",
    "    df.columns = [\"num\", \"percent\"]\n",
    "    return df.sort_values(by=\"percent\", ascending=False)\n",
    "\n",
    "\n",
    "df_labels_counts = create_value_counts_df(df, col=label_col)\n",
    "df_labels_counts"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can see that the distribution of the classes is uneven.    \n",
    "Our class distribution is apporximately  60%, 25%, 15% between the lived, died, euthanized classes (respectively).   \n",
    "\n",
    "**But how are our NaNs distriputed**?     \n",
    "\n",
    "What is the distribution of `NaN`s in each feature with relation to our classification field.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>all_sum_na</th>\n",
       "      <th>lived_sum_na</th>\n",
       "      <th>died_sum_na</th>\n",
       "      <th>euthanized_sum_na</th>\n",
       "      <th>all_percentage_na</th>\n",
       "      <th>lived_percentage_na</th>\n",
       "      <th>died_percentage_na</th>\n",
       "      <th>euthanized_percentage_na</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>rectal_temperature</th>\n",
       "      <td>60</td>\n",
       "      <td>26</td>\n",
       "      <td>24</td>\n",
       "      <td>10</td>\n",
       "      <td>0.20</td>\n",
       "      <td>0.15</td>\n",
       "      <td>0.31</td>\n",
       "      <td>0.23</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>pulse</th>\n",
       "      <td>24</td>\n",
       "      <td>12</td>\n",
       "      <td>11</td>\n",
       "      <td>1</td>\n",
       "      <td>0.08</td>\n",
       "      <td>0.07</td>\n",
       "      <td>0.14</td>\n",
       "      <td>0.02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>respiratory_rate</th>\n",
       "      <td>58</td>\n",
       "      <td>31</td>\n",
       "      <td>19</td>\n",
       "      <td>8</td>\n",
       "      <td>0.19</td>\n",
       "      <td>0.17</td>\n",
       "      <td>0.25</td>\n",
       "      <td>0.18</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>temperature_of_extremities</th>\n",
       "      <td>56</td>\n",
       "      <td>32</td>\n",
       "      <td>13</td>\n",
       "      <td>11</td>\n",
       "      <td>0.19</td>\n",
       "      <td>0.18</td>\n",
       "      <td>0.17</td>\n",
       "      <td>0.25</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>peripheral_pulse</th>\n",
       "      <td>69</td>\n",
       "      <td>39</td>\n",
       "      <td>18</td>\n",
       "      <td>12</td>\n",
       "      <td>0.23</td>\n",
       "      <td>0.22</td>\n",
       "      <td>0.23</td>\n",
       "      <td>0.27</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mucous_membranes</th>\n",
       "      <td>47</td>\n",
       "      <td>28</td>\n",
       "      <td>11</td>\n",
       "      <td>8</td>\n",
       "      <td>0.16</td>\n",
       "      <td>0.16</td>\n",
       "      <td>0.14</td>\n",
       "      <td>0.18</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>capillary_refill_time</th>\n",
       "      <td>32</td>\n",
       "      <td>19</td>\n",
       "      <td>10</td>\n",
       "      <td>3</td>\n",
       "      <td>0.11</td>\n",
       "      <td>0.11</td>\n",
       "      <td>0.13</td>\n",
       "      <td>0.07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>pain</th>\n",
       "      <td>55</td>\n",
       "      <td>34</td>\n",
       "      <td>12</td>\n",
       "      <td>9</td>\n",
       "      <td>0.18</td>\n",
       "      <td>0.19</td>\n",
       "      <td>0.16</td>\n",
       "      <td>0.20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>peristalsis</th>\n",
       "      <td>44</td>\n",
       "      <td>22</td>\n",
       "      <td>15</td>\n",
       "      <td>7</td>\n",
       "      <td>0.15</td>\n",
       "      <td>0.12</td>\n",
       "      <td>0.19</td>\n",
       "      <td>0.16</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abdominal_distension</th>\n",
       "      <td>56</td>\n",
       "      <td>31</td>\n",
       "      <td>14</td>\n",
       "      <td>11</td>\n",
       "      <td>0.19</td>\n",
       "      <td>0.17</td>\n",
       "      <td>0.18</td>\n",
       "      <td>0.25</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>nasogastric_tube</th>\n",
       "      <td>104</td>\n",
       "      <td>62</td>\n",
       "      <td>25</td>\n",
       "      <td>17</td>\n",
       "      <td>0.35</td>\n",
       "      <td>0.35</td>\n",
       "      <td>0.32</td>\n",
       "      <td>0.39</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>rectal_examination</th>\n",
       "      <td>102</td>\n",
       "      <td>56</td>\n",
       "      <td>26</td>\n",
       "      <td>20</td>\n",
       "      <td>0.34</td>\n",
       "      <td>0.31</td>\n",
       "      <td>0.34</td>\n",
       "      <td>0.45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>packed_cell_volume</th>\n",
       "      <td>29</td>\n",
       "      <td>13</td>\n",
       "      <td>8</td>\n",
       "      <td>8</td>\n",
       "      <td>0.10</td>\n",
       "      <td>0.07</td>\n",
       "      <td>0.10</td>\n",
       "      <td>0.18</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>total_protein</th>\n",
       "      <td>33</td>\n",
       "      <td>13</td>\n",
       "      <td>12</td>\n",
       "      <td>8</td>\n",
       "      <td>0.11</td>\n",
       "      <td>0.07</td>\n",
       "      <td>0.16</td>\n",
       "      <td>0.18</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                            all_sum_na  lived_sum_na  died_sum_na  \\\n",
       "rectal_temperature                  60            26           24   \n",
       "pulse                               24            12           11   \n",
       "respiratory_rate                    58            31           19   \n",
       "temperature_of_extremities          56            32           13   \n",
       "peripheral_pulse                    69            39           18   \n",
       "mucous_membranes                    47            28           11   \n",
       "capillary_refill_time               32            19           10   \n",
       "pain                                55            34           12   \n",
       "peristalsis                         44            22           15   \n",
       "abdominal_distension                56            31           14   \n",
       "nasogastric_tube                   104            62           25   \n",
       "rectal_examination                 102            56           26   \n",
       "packed_cell_volume                  29            13            8   \n",
       "total_protein                       33            13           12   \n",
       "\n",
       "                            euthanized_sum_na  all_percentage_na  \\\n",
       "rectal_temperature                         10               0.20   \n",
       "pulse                                       1               0.08   \n",
       "respiratory_rate                            8               0.19   \n",
       "temperature_of_extremities                 11               0.19   \n",
       "peripheral_pulse                           12               0.23   \n",
       "mucous_membranes                            8               0.16   \n",
       "capillary_refill_time                       3               0.11   \n",
       "pain                                        9               0.18   \n",
       "peristalsis                                 7               0.15   \n",
       "abdominal_distension                       11               0.19   \n",
       "nasogastric_tube                           17               0.35   \n",
       "rectal_examination                         20               0.34   \n",
       "packed_cell_volume                          8               0.10   \n",
       "total_protein                               8               0.11   \n",
       "\n",
       "                            lived_percentage_na  died_percentage_na  \\\n",
       "rectal_temperature                         0.15                0.31   \n",
       "pulse                                      0.07                0.14   \n",
       "respiratory_rate                           0.17                0.25   \n",
       "temperature_of_extremities                 0.18                0.17   \n",
       "peripheral_pulse                           0.22                0.23   \n",
       "mucous_membranes                           0.16                0.14   \n",
       "capillary_refill_time                      0.11                0.13   \n",
       "pain                                       0.19                0.16   \n",
       "peristalsis                                0.12                0.19   \n",
       "abdominal_distension                       0.17                0.18   \n",
       "nasogastric_tube                           0.35                0.32   \n",
       "rectal_examination                         0.31                0.34   \n",
       "packed_cell_volume                         0.07                0.10   \n",
       "total_protein                              0.07                0.16   \n",
       "\n",
       "                            euthanized_percentage_na  \n",
       "rectal_temperature                              0.23  \n",
       "pulse                                           0.02  \n",
       "respiratory_rate                                0.18  \n",
       "temperature_of_extremities                      0.25  \n",
       "peripheral_pulse                                0.27  \n",
       "mucous_membranes                                0.18  \n",
       "capillary_refill_time                           0.07  \n",
       "pain                                            0.20  \n",
       "peristalsis                                     0.16  \n",
       "abdominal_distension                            0.25  \n",
       "nasogastric_tube                                0.39  \n",
       "rectal_examination                              0.45  \n",
       "packed_cell_volume                              0.18  \n",
       "total_protein                                   0.18  "
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def get_na_per_label(df, label_col, labels=df_labels_counts.index):\n",
    "    cols = [c for c in df.columns if c != label_col]\n",
    "    sum_na = (\n",
    "        [df[col].isna().sum()]\n",
    "        + [\n",
    "            df[df[label_col] == label][col].isna().sum()\n",
    "            for label in df_labels_counts.index.tolist()\n",
    "        ]\n",
    "        for col in cols\n",
    "    )\n",
    "    df_sum_na = pd.DataFrame(\n",
    "        columns=[f\"{col}_sum_na\" for col in [\"all\"] + df_labels_counts.index.tolist()],\n",
    "        data=sum_na,\n",
    "        index=cols,\n",
    "    )\n",
    "    df_sum_na[\"all_percentage_na\"] = df.isna().sum() / len(df)\n",
    "    for label in labels:\n",
    "        df_sum_na[f\"{label}_percentage_na\"] = (\n",
    "            df_sum_na[f\"{label}_sum_na\"] / df_labels_counts.loc[label, \"num\"]\n",
    "        )\n",
    "    return df_sum_na\n",
    "\n",
    "def get_na_cols(df):\n",
    "    _s_na = df.isna().sum()\n",
    "    na_cols = _s_na[_s_na > 0].index\n",
    "    na_cols = list(na_cols) + [label_col]\n",
    "    return na_cols\n",
    "\n",
    "na_cols = get_na_cols(df_nzr_threshold)\n",
    "df_sum_na = get_na_per_label(df_nzr_threshold[na_cols], label_col)\n",
    "df_sum_na"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can see that the distributions of the `NaN`s across the classification field are not even. _e.g._ the `rectal_temperature` feature has twice as much NaN in the `died` & `lived` classes than in the `euthanized` class.  \n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Assuming that we don't want to remove any features that have less than 15% of `NaN`s, no matter how the `NaN` distribution is across the classification field."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>all_sum_na</th>\n",
       "      <th>lived_sum_na</th>\n",
       "      <th>died_sum_na</th>\n",
       "      <th>euthanized_sum_na</th>\n",
       "      <th>all_percentage_na</th>\n",
       "      <th>lived_percentage_na</th>\n",
       "      <th>died_percentage_na</th>\n",
       "      <th>euthanized_percentage_na</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>rectal_temperature</th>\n",
       "      <td>60</td>\n",
       "      <td>26</td>\n",
       "      <td>24</td>\n",
       "      <td>10</td>\n",
       "      <td>0.20</td>\n",
       "      <td>0.15</td>\n",
       "      <td>0.31</td>\n",
       "      <td>0.23</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>respiratory_rate</th>\n",
       "      <td>58</td>\n",
       "      <td>31</td>\n",
       "      <td>19</td>\n",
       "      <td>8</td>\n",
       "      <td>0.19</td>\n",
       "      <td>0.17</td>\n",
       "      <td>0.25</td>\n",
       "      <td>0.18</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>temperature_of_extremities</th>\n",
       "      <td>56</td>\n",
       "      <td>32</td>\n",
       "      <td>13</td>\n",
       "      <td>11</td>\n",
       "      <td>0.19</td>\n",
       "      <td>0.18</td>\n",
       "      <td>0.17</td>\n",
       "      <td>0.25</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>peripheral_pulse</th>\n",
       "      <td>69</td>\n",
       "      <td>39</td>\n",
       "      <td>18</td>\n",
       "      <td>12</td>\n",
       "      <td>0.23</td>\n",
       "      <td>0.22</td>\n",
       "      <td>0.23</td>\n",
       "      <td>0.27</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mucous_membranes</th>\n",
       "      <td>47</td>\n",
       "      <td>28</td>\n",
       "      <td>11</td>\n",
       "      <td>8</td>\n",
       "      <td>0.16</td>\n",
       "      <td>0.16</td>\n",
       "      <td>0.14</td>\n",
       "      <td>0.18</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>pain</th>\n",
       "      <td>55</td>\n",
       "      <td>34</td>\n",
       "      <td>12</td>\n",
       "      <td>9</td>\n",
       "      <td>0.18</td>\n",
       "      <td>0.19</td>\n",
       "      <td>0.16</td>\n",
       "      <td>0.20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abdominal_distension</th>\n",
       "      <td>56</td>\n",
       "      <td>31</td>\n",
       "      <td>14</td>\n",
       "      <td>11</td>\n",
       "      <td>0.19</td>\n",
       "      <td>0.17</td>\n",
       "      <td>0.18</td>\n",
       "      <td>0.25</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>nasogastric_tube</th>\n",
       "      <td>104</td>\n",
       "      <td>62</td>\n",
       "      <td>25</td>\n",
       "      <td>17</td>\n",
       "      <td>0.35</td>\n",
       "      <td>0.35</td>\n",
       "      <td>0.32</td>\n",
       "      <td>0.39</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>rectal_examination</th>\n",
       "      <td>102</td>\n",
       "      <td>56</td>\n",
       "      <td>26</td>\n",
       "      <td>20</td>\n",
       "      <td>0.34</td>\n",
       "      <td>0.31</td>\n",
       "      <td>0.34</td>\n",
       "      <td>0.45</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                            all_sum_na  lived_sum_na  died_sum_na  \\\n",
       "rectal_temperature                  60            26           24   \n",
       "respiratory_rate                    58            31           19   \n",
       "temperature_of_extremities          56            32           13   \n",
       "peripheral_pulse                    69            39           18   \n",
       "mucous_membranes                    47            28           11   \n",
       "pain                                55            34           12   \n",
       "abdominal_distension                56            31           14   \n",
       "nasogastric_tube                   104            62           25   \n",
       "rectal_examination                 102            56           26   \n",
       "\n",
       "                            euthanized_sum_na  all_percentage_na  \\\n",
       "rectal_temperature                         10               0.20   \n",
       "respiratory_rate                            8               0.19   \n",
       "temperature_of_extremities                 11               0.19   \n",
       "peripheral_pulse                           12               0.23   \n",
       "mucous_membranes                            8               0.16   \n",
       "pain                                        9               0.18   \n",
       "abdominal_distension                       11               0.19   \n",
       "nasogastric_tube                           17               0.35   \n",
       "rectal_examination                         20               0.34   \n",
       "\n",
       "                            lived_percentage_na  died_percentage_na  \\\n",
       "rectal_temperature                         0.15                0.31   \n",
       "respiratory_rate                           0.17                0.25   \n",
       "temperature_of_extremities                 0.18                0.17   \n",
       "peripheral_pulse                           0.22                0.23   \n",
       "mucous_membranes                           0.16                0.14   \n",
       "pain                                       0.19                0.16   \n",
       "abdominal_distension                       0.17                0.18   \n",
       "nasogastric_tube                           0.35                0.32   \n",
       "rectal_examination                         0.31                0.34   \n",
       "\n",
       "                            euthanized_percentage_na  \n",
       "rectal_temperature                              0.23  \n",
       "respiratory_rate                                0.18  \n",
       "temperature_of_extremities                      0.25  \n",
       "peripheral_pulse                                0.27  \n",
       "mucous_membranes                                0.18  \n",
       "pain                                            0.20  \n",
       "abdominal_distension                            0.25  \n",
       "nasogastric_tube                                0.39  \n",
       "rectal_examination                              0.45  "
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "threshold_min_na = 0.15\n",
    "mask_threshold_min_na = df_sum_na[\"all_percentage_na\"] > threshold_min_na\n",
    "df_sum_min_na = df_sum_na[mask_threshold_min_na].copy()\n",
    "df_sum_min_na"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we can further analyse our data. Let's see the ratio between the classifications."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "def create_ratio_between_classes(df, label_classes):\n",
    "    for label_a, label_b in combinations(label_classes, 2):\n",
    "        df[f\"ratio_percentage_{label_a}_{label_b}\"] = (\n",
    "            df[f\"{label_a}_percentage_na\"] / df[f\"{label_b}_percentage_na\"]\n",
    "        )\n",
    "    col_ratio = [col for col in df.columns if \"ratio_percentage\" in col]\n",
    "    return df[col_ratio]\n",
    "\n",
    "\n",
    "create_ratio_between_classes(df_sum_min_na, df_labels_counts.index).plot(\n",
    "    figsize=(12, 5), rot=45, grid=True\n",
    ");"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Values closer to 1 have similar percentage of `NaN`s, whereas values that are further away a higher distributions of the `NaN`s across the classification field.  \n",
    "\n",
    "We can set a lower and upper threshold for filtering out the problematic features. Once all the ratios are between these limits we will want to keep this feature. Any value outside these limits we can assume that the `NaN`s are unevenly distributed, and the features should be removed. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['rectal_examination',\n",
       " 'abdominal_distension',\n",
       " 'rectal_temperature',\n",
       " 'temperature_of_extremities',\n",
       " 'respiratory_rate']"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def get_features_outside_threshold(\n",
    "    df,\n",
    "    lt_ratio_threshold=0.7,\n",
    "    gt_ratio_threshold=1.3,\n",
    "):\n",
    "    features_to_drop = []\n",
    "    for col in [col for col in df.columns if \"ratio\" in col]:\n",
    "        mask_ratio_threshold = ~(\n",
    "            df[col].between(lt_ratio_threshold, gt_ratio_threshold)\n",
    "        )\n",
    "        features_to_drop.extend(mask_ratio_threshold[mask_ratio_threshold].index)\n",
    "    features_to_drop = list(set(features_to_drop))\n",
    "    return features_to_drop\n",
    "\n",
    "\n",
    "features_to_drop = get_features_outside_threshold(df_sum_min_na)\n",
    "features_to_drop"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "After removing the features from get_features_outside_threshold function we are left with 9 features with `NaN`s that we are going to imputate\n"
     ]
    }
   ],
   "source": [
    "df_for_model = df_nzr_threshold.drop(columns=features_to_drop)\n",
    "print(\n",
    "    f\"After removing the features from get_features_outside_threshold function we are left with {number_of_features_with_NaN(df_for_model)} features with `NaN`s that we are going to imputate\"\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Summary  \n",
    "\n",
    "This post describes the issues while analysing `NaN`s for feature selection.  \n",
    "\n",
    "Simple filtering methods do not always perform as expected and additional emphsesis should be taken when working with sparse matrices.\n",
    "\n",
    "We can analyse `NaN` within features in multiple levels:  \n",
    "1. At the global level - i.e. the total amount of NaNs within a feature  (both for removing and for keeping features) \n",
    "2. At the label/classification level - i.e. the relative distribution of NaNs per class.\n",
    "\n",
    "Finally we recommand trying out the [missingno package](https://github.com/ResidentMario/missingno) for graphical analysis of `NaN` values\n"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}